Describes a scaled down AIXI agent that uses Monte Carlo Tree Search.

Shane's thesis. Good intro to Bayesian sequence prediction, Solomonoff's universal prior and Solomonoff induction.

Suggests Monte Carlo sampling of programs according to their (time-bound) algorithmic probability as an approximation for the optimal but incomputable AIXI theory. A few thoughts on how our environment should bias choice of hypotheses for induction.

The AIXI model is the most intelligent unbiased agent possible (though uncomputable). A modified version, AIXItl, is still effectively more intelligent than any other time t and space l bounded agent.